Development of a digital twin framework for hybrid adsorption-ultrafiltration systems in drinking water treatment

A data-driven digital twin was developed to address the ongoing challenges of humic acid removal and membrane fouling in hybrid adsorption-ultrafiltration (UF) systems used for drinking water treatment. Natural organic matter, particularly humic substances, continues to pose operational challenges in membrane-based processes, leading to irreversible fouling and flux reduction. The digital twin incorporates eXtreme Gradient Boosting (XGBoost) models combined with SHapley Additive exPlanations (SHAP) to forecast key performance indicators such as dissolved organic carbon (DOC), UV254 absorbance, specific UV absorbance (SUVA), and membrane fouling percentage under various hydraulic and chemical loading conditions. The UF-only model showed the highest accuracy, with R2 values of 0.98 for DOC, 0.99 for UV254, 0.80 for SUVA, and 0.99 for fouling. The hybrid GAC-UF model also performed strongly in predicting fouling (R2 = 0.99) and UV254 (R2= 0.79), with moderate skill in DOC removal. These models are integrated into a forward-simulation framework that enables real-time scenario testing, allowing operators to assess system responses and receive guidance on filtration cycle limits and pretreatment effectiveness. The digital twin offers a reliable decision-support platform suitable for advanced treatment systems. It improves understanding of process dynamics and provides transparent, interpretable insights into operational sensitivities, supporting informed decision-making. This framework paves the way for AI-driven optimization and predictive control in modern water treatment facilities.